A multi-wavelength lidar aerosol microphysical characteristics inversion method and system

Abstract

The present invention proposes a multi-wavelength lidar aerosol microphysical property inversion method and system. The present invention constructs a normalized lookup table and normalized input based on Mie theory calculation and normalization method; based on the k-nearest neighbor theory, a reduced solution space is obtained according to the optical property gap preservation method; the gap between the normalized input and each normalized optical property in the reduced solution space is calculated, and a first set of microphysical property solutions is further calculated using a sorting decision tree method; a constraint window is constructed through optical property normalization and k-nearest neighbor theory; based on the k-nearest neighbor theory, a refined solution space is obtained by reducing and interpolating the optical property gap preservation method and the constraint window; a final microphysical property solution and a final optical property solution are obtained using a sorting decision tree method, and the remaining microphysical properties are calculated. The present invention improves the stability of the inversion, effectively overcomes the ill-posedness of the inversion problem, and improves the inversion accuracy.

Description

Technical Field

[0001] The present invention belongs to the field of laser radar measurement technology, and in particular relates to a multi-wavelength laser radar aerosol microphysical property inversion method and system. Background Art

[0002] Atmospheric aerosols, solid or liquid particles suspended in the atmosphere, play a significant role in Earth's climate, radiation, and environment. First, aerosols absorb and scatter solar radiation, affecting global and regional climate by altering the radiation budget. Second, aerosols are a key contributor to pollution phenomena such as haze, which has direct adverse effects on human health. Finally, aerosols provide condensation nuclei for the condensation or freezing of atmospheric water vapor, influencing the water cycle and forming clouds and precipitation. Aerosols typically exhibit a bimodal size distribution, consisting of fine and coarse modes. Aerosols with a predominant fine mode fraction in the size distribution are considered fine mode aerosols, such as urban pollution particles and smog. Therefore, fine mode aerosols are closely related to human activities and are a key research topic in climate prediction. To better quantify the impact of fine mode aerosols, it is necessary to study their optical and microphysical properties, such as the particle size distribution (APSD), complex refractive index (CRI), single scattering reflectance (SSA), and concentration parameters. Because of the spatiotemporal variability of aerosol properties, its vertical distribution information becomes the key to accurately understand the complex effects of aerosols.

[0003] LiDAR is the only active remote sensing technology for detecting vertical aerosol information. Multi-wavelength lidars, such as Raman lidar and high-spectral-resolution lidar, can provide backscattering coefficients (β) and extinction coefficients (α) at ​​multiple wavelengths and have been demonstrated to be capable of retrieving aerosol microphysical properties. Numerous mature algorithms have been developed or implemented for the well-known 3β+2α configuration, which combines backscattering coefficients at 355nm, 532nm, and 1064nm with extinction coefficients at 355nm and 532nm. These algorithms, such as principal component analysis, regularization, and linear estimation, have been developed and are in use. However, due to the limited information provided by the input data, the inversion system is ill-posed. These methods often introduce additional constraints, require manual supervision, or are extremely time-consuming. Furthermore, they place high demands on the lidar system and rely primarily on the 3β+2α configuration input, making it difficult to guarantee inversion results for systems lacking some channels. The latest solution was proposed by Liu Dong's team at Zhejiang University. This algorithm combines k-nearest neighbor and random forest theory to complete the unsupervised and efficient inversion of large amounts of lidar data, but its stability and accuracy still need to be improved. Summary of the Invention

[0004] In view of the shortcomings of the existing multi-wavelength lidar aerosol microphysical properties inversion method, the present invention proposes a multi-wavelength lidar aerosol microphysical properties inversion method and system to solve the problems in the existing technology that the inversion results are unstable, inaccurate, inefficient, and not suitable for small-channel lidar systems.

[0005] The technical solution of the method of the present invention is a multi-wavelength lidar aerosol microphysical property inversion method, comprising the following steps:

[0006] Step 1: Based on the Mie theory, the aerosol lognormal particle size distribution model is used to calculate the backscattering coefficient at each backscattering wavelength and the extinction coefficient at each extinction wavelength. A lookup table is then established. The optical characteristics of the lookup table and the input of the multi-wavelength lidar are normalized using a normalization method to obtain a normalized lookup table and normalized input.

[0007] Step 2: Based on the k-nearest neighbor theory, the gap between the normalized input and each group of normalized optical properties in the normalized lookup table is calculated and sorted from small to large, and the elements of the groups with the smallest gap are retained to obtain a narrowed solution space;

[0008] Step 3: Calculate the gap between the normalized input and each normalized optical property in the reduced solution space, sort the normalized optical properties from large to small based on the gap between the normalized input and each normalized optical property in the reduced solution space, and generate decision trees with different pruning orders based on the weighted sorting. Prune each decision tree according to the included pruning order to generate possible solutions. Average the possible solutions of all decision trees to obtain the first set of microphysical property solutions.

[0009] Step 4: Calculate the corresponding optical properties based on the first set of microphysical property solutions and normalize them to obtain the first set of optical property solutions. Based on the k-nearest neighbor theory, calculate the gap between the union of the first set of microphysical property solutions and the first set of optical property solutions and each element in the normalized lookup table, and retain the elements in the groups with the smallest gap to obtain the constraint window.

[0010] Step 5: Based on the k-nearest neighbor theory, follow the method in step 2 and use the constraint window instead of the normalized lookup table. Reduce the constraint window and perform interpolation to obtain the refined solution space.

[0011] Step 6: Following the method of step 3, the refined solution space is used instead of the reduced solution space to obtain the final solution of the microphysical properties and the final solution of the optical properties, and the remaining microphysical properties are calculated;

[0012] Preferably, in step 1, the backscattering coefficient at each backscattering wavelength and the extinction coefficient at each extinction wavelength are calculated separately using the aerosol lognormal particle size distribution model according to Mie theory, and a lookup table is established, specifically as follows:

[0013] Input the real part of the complex refractive index, the imaginary part of the complex refractive index, the mode radius and the logarithmic geometric variance in the aerosol microphysical properties, and calculate in sequence the backscattering coefficient at each backscattering wavelength according to the Mie theory through the aerosol lognormal particle size distribution model according to each backscattering wavelength. Combined with each extinction wavelength, calculate in sequence the extinction coefficient at each extinction wavelength according to the Mie theory through the aerosol lognormal particle size distribution model; according to the number of groups N LUT The real part of the complex refractive index, the imaginary part of the complex refractive index, the mode radius and the logarithmic geometric variance are used to calculate the corresponding backscattering coefficient and extinction coefficient, and a lookup table is constructed together; the lookup table contains N LUT Group elements, each group includes the real part of the complex refractive index, the imaginary part of the complex refractive index, the mode radius and the logarithmic geometric variance as well as the corresponding backscattering coefficient and extinction coefficient;

[0014] The combined normalization method is used to normalize the lookup table optical characteristics and the input of the multi-wavelength lidar, as follows:

[0015]

[0016]

[0017]

[0018] N norm =N+K+N×K

[0019] Among them, β k represents the backscattering coefficient at the kth backscattering wavelength, B k represents the normalized backscattering coefficient at the kth backscattering wavelength, B norm represents the second-order norm of the backscattering coefficient at all backscattering wavelengths, K represents the number of backscattering wavelengths; α n represents the extinction coefficient at the nth extinction wavelength, A n represents the normalized extinction coefficient at the nth extinction wavelength, A norm represents the second-order norm of the optical coefficient at all extinction wavelengths, and N represents the number of extinction wavelengths; Represents the extinction backscattering ratio between the kth backscattering wavelength and the nth extinction wavelength; the normalized optical properties include the normalized backscattering coefficient, the normalized extinction coefficient and the extinction backscattering ratio, N norm represents the normalized optical property quantity;

[0020] Preferably, the step 2 of calculating the difference between the normalized input and each set of normalized optical properties in the normalized lookup table is as follows:

[0021]

[0022] Among them, Dist M (n) represents the difference between each set of normalized optical properties in the normalized lookup table and the normalized input, represents the normalized optical properties in the normalized lookup table, n represents the nth group element, G input represents the normalized input, S represents With G input The covariance matrix of

[0023] Step 2 sorts the elements in ascending order and retains the elements with the smallest gap, as follows:

[0024] N kNN =ω1*N LUT

[0025] where N kNN Represents the number of element groups retained, and ω1 represents the ratio of the number of groups retained when generating the reduced solution space.

[0026] Preferably, the calculation of the normalized input in step 3 and the reduction of the gap between each normalized optical property in the solution space are as follows:

[0027]

[0028] Among them, Dist opt (i) represents the gap between the normalized input and each normalized optical property in the reduced solution space, represents the normalized optical properties in the reduced solution space, i represents different normalized optical properties;

[0029] In step 3, the normalized optical properties are sorted from large to small based on the gap between the normalized input and each normalized optical property in the solution space, and a decision tree with different pruning orders is generated based on the weighted sorting, as follows:

[0030]

[0031] Among them, P(i) represents the probability of the i-th normalized optical property being extracted after sorting; the process of generating decision trees with different pruning orders is: the total number of generated trees is N RF For each decision tree, all normalized optical properties are extracted without replacement according to P(i) to generate a sequence of normalized optical properties as the pruning order;

[0032] Each decision tree described in step 3 is pruned according to the included pruning order to generate possible solutions, as follows:

[0033] N t(i+1)=N t (i)×(1-ω2)

[0034] 0≤i≤N norm -1,N t (0) = N kNN

[0035] where N t (i) represents the number of element groups retained after the i-th pruning, ω2 represents the pruning coefficient, N t (0) represents the initial number of element groups.

[0036] Preferably, in step 4, based on the k-nearest neighbor theory, the gap between the union of the first set of microphysical property solutions and the first set of optical property solutions and each set of elements in the normalized lookup table is calculated, as follows:

[0037]

[0038] Among them, Dist M,ret (n) represents the difference between the first set of microphysical property solutions and the first set of optical property solutions and each element of the normalized lookup table, G LUT (n) represents the normalized lookup table, n represents the nth group element of the normalized lookup table, G ret S represents the union of the first set of microphysical property solutions and the first set of optical property solutions G LUT (n) and G ret The covariance matrix of

[0039] Step 4 sorts the elements in ascending order and retains the elements with the smallest gap, as follows:

[0040] N C =ω3*N LUT

[0041] where N C Represents the number of element groups retained, and ω3 represents the ratio of the number of groups retained by the generated constraint window.

[0042] Preferably, the interpolation operation is performed on the reduced constraint window in step 5, specifically as follows:

[0043] Fixing three of the real part of the complex refractive index, the imaginary part of the complex refractive index, the mode radius and the logarithmic geometric variance, extract all the values ​​of the remaining microphysical properties and their corresponding optical properties in the reduced constraint window, a total of N norm The same processing is performed on the remaining three of the real part of the complex refractive index, the imaginary part of the complex refractive index, the mode radius and the logarithmic geometric variance to obtain the refined solution space; define P inter is the interpolation ratio, each P interInterpolate between data points.

[0044] Preferably, the remaining microphysical properties are calculated as follows:

[0045]

[0046]

[0047]

[0048]

[0049] in, represents the volume concentration obtained by inversion, represents the backscattering coefficient at the input k-th backscattering wavelength, represents the extinction coefficient at the input nth extinction wavelength, represents the backscattering coefficient at the kth backscattering wavelength obtained by inversion, Indicates the extinction coefficient at the nth extinction wavelength obtained by inversion, which can be calculated according to the method in step 1, N λ is the sum of the backscattering wavelength and the extinction wavelength;

[0050] v ret (r) represents the volume particle size spectrum distribution obtained by inversion, lnσ ret represents the logarithmic geometric variance obtained by inversion, represents the pattern radius obtained by inversion, represents the effective particle size obtained by inversion, Indicates the inverted wavelength λ x The single scattering albedo at , represents the real part of the complex refractive index obtained by inversion, represents the imaginary part of the complex refractive index obtained by inversion, and Represents wavelength λ x The scattering coefficient and extinction coefficient at can be calculated by Mie theory.

[0051] The present invention also provides a multi-wavelength lidar aerosol microphysical property inversion system, comprising:

[0052] The normalized lookup table and normalized input construction modules are used to calculate the backscattering coefficient at each backscattering wavelength and the extinction coefficient at each extinction wavelength based on the Mie theory using the aerosol lognormal particle size distribution model. The normalized lookup table is then constructed using the normalization method to obtain the normalized input.

[0053] A solution space reduction calculation module is used to calculate the gap between the normalized input and each group of normalized optical properties in the normalized lookup table based on the k-nearest neighbor theory, sort them from small to large, and retain the group elements with the smallest gap to obtain a reduced solution space;

[0054] A first set of microphysical property solution solving modules is configured to calculate the difference between the normalized input and each normalized optical property in the reduced solution space, sort the normalized optical properties from largest to smallest based on the difference between the normalized input and each normalized optical property in the reduced solution space, generate decision trees with different pruning orders based on the weighted sorting, prune each decision tree according to the included pruning order, generate possible solutions, and average the possible solutions of all decision trees to obtain the first set of microphysical property solutions;

[0055] A constraint window construction module is used to calculate the corresponding optical properties based on the first set of microphysical property solutions and normalize them to obtain the first set of optical property solutions. Based on the k-nearest neighbor theory, the module calculates the gap between the union of the first set of microphysical property solutions and the first set of optical property solutions and each set of elements in the normalized lookup table, and retains the elements of the groups with the smallest gap to obtain the constraint window.

[0056] A refined solution space construction module is used to obtain a refined solution space based on the k-nearest neighbor theory, through the first set of microphysical characteristic solution solving modules, using a constraint window instead of a normalized lookup table, reducing the constraint window, and performing interpolation operations;

[0057] The final solution solving module is combined with the first set of microphysical characteristic solution solving modules, and uses the refined solution space instead of the reduced solution space to obtain the final solution of the microphysical characteristics and the final solution of the optical characteristics, and calculate the remaining microphysical characteristics;

[0058] Compared with the existing technology, the advantages of the method proposed in the present invention are: the present invention optimizes the generation scheme of the decision tree in the random forest process, which can improve the stability of the inversion; the present invention improves the inversion process in a feedback iterative manner, which can effectively overcome the ill-posedness of the inversion problem and improve the inversion accuracy; while ensuring the inversion effect of the 3β+2α configuration, the present invention improves the inversion effect of the 3β+1α and 2β+1α configurations, can provide application conditions for simple lidar systems, reduce the cost of lidar detection tasks, and is of great significance to the use of lidar for aerosol detection. BRIEF DESCRIPTION OF THE DRAWINGS

[0059] Figure 1 : A flow chart of a method according to an embodiment of the present invention;

[0060] Figure 2 : The result of inverting the error-free data in the embodiment of the present invention;

[0061] Figure 3: The results of repeated inversion of the same data under the 3β+2α configuration in the embodiment of the present invention. DETAILED DESCRIPTION

[0062] To make the purpose, technical solutions, and advantages of the embodiments of this application more clear, the technical solutions in the embodiments of this application will be clearly and completely described below in conjunction with the drawings in the embodiments of this application. Obviously, the described embodiments are part of the embodiments of this application, not all of the embodiments. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of this application.

[0063] The following is combined with Figure 1-3 The embodiments of the present invention are described in further detail.

[0064] like Figure 1 As shown, the method flow chart of the embodiment of the present invention includes:

[0065] Step 1: Input the real part of the complex refractive index, the imaginary part of the complex refractive index, the mode radius, and the logarithmic geometric variance of the aerosol microphysical properties, and calculate the backscattering coefficient at each backscattering wavelength in sequence according to the Mie theory using the aerosol lognormal particle size distribution model. Combined with each extinction wavelength, calculate the extinction coefficient at each extinction wavelength in sequence according to the Mie theory using the aerosol lognormal particle size distribution model. Use multiple sets of the real part of the complex refractive index, the imaginary part of the complex refractive index, the mode radius, and the logarithmic geometric variance and the calculated backscattering coefficients and extinction coefficients to establish a lookup table. Normalize the backscattering coefficient and extinction coefficient from the multi-wavelength lidar input and the backscattering coefficient and extinction coefficient in the lookup table to obtain normalized input and normalized lookup table.

[0066] In step 1, the backscattering coefficient at each backscattering wavelength is normalized as follows:

[0067]

[0068]

[0069] Among them, β k represents the backscattering coefficient at the kth backscattering wavelength, B k represents the normalized backscattering coefficient at the kth backscattering wavelength, B norm represents the second-order norm of the backscattering coefficient at all backscattering wavelengths, and K represents the number of backscattering wavelengths;

[0070] In step 1, the extinction coefficient at each extinction wavelength is normalized as follows:

[0071]

[0072]

[0073] Among them, α n represents the extinction coefficient at the nth extinction wavelength, A n represents the normalized extinction coefficient at the nth extinction wavelength, A norm represents the second-order norm of the optical coefficient at all extinction wavelengths, and N represents the number of extinction wavelengths;

[0074] The extinction backscattering ratio between each backscattering wavelength and each extinction wavelength is calculated as described in step 1 as follows:

[0075]

[0076] in, It represents the extinction backscattering ratio between the kth backscattering wavelength and the nth extinction wavelength;

[0077] The normalized optical characteristic quantities are calculated as described in step 1 as follows:

[0078] N norm =N+K+N×K

[0079] Among them, N norm represents the number of normalized optical properties, K represents the number of backscattered wavelengths, and N represents the number of extinction wavelengths;

[0080] The lookup table described in step 1 is constructed using multiple sets of real part of complex refractive index, imaginary part of complex refractive index, mode radius and logarithmic geometric variance and the backscattering coefficient and extinction coefficient calculated therefrom, as follows:

[0081] According to the number of groups N LUT The real part of the complex refractive index, the imaginary part of the complex refractive index, the mode radius and the logarithmic geometric variance are used to calculate the corresponding backscattering coefficient and extinction coefficient, and a lookup table is constructed together; the lookup table contains N LUT Group elements, each group includes the real part of the complex refractive index, the imaginary part of the complex refractive index, the mode radius and the logarithmic geometric variance as well as the corresponding backscattering coefficient and extinction coefficient;

[0082] In this embodiment, a 3β+2α configuration including backscattering coefficients at 355, 532, and 1064 nm and extinction coefficients at 355 and 532 nm, a 3β+1α configuration including backscattering coefficients at 355, 532, and 1064 nm and extinction coefficients at 355 nm, and a 2β+1α configuration including backscattering coefficients at 532 and 1064 nm and extinction coefficients at 355 nm are selected; the microphysical property combination for constructing the lookup table is: m r=1.30:0.02:1.70, m i =0.00:0.001:0.05, r med =50:10:500nm, lnσ=0.38:0.01:0.50, so N LUT =640458. In order to verify the performance of the method, this embodiment uses multiple sets of microphysical characteristics to calculate the optical characteristics as simulation inputs to simulate the method. The combination of microphysical characteristics simulation input is: m r =1.35,1.45,1.55,1.65,m i =0.001,0.005,0.01,0.015,0.020,0.025,r med =70, 100, 140, 180, 240, 300 nm, lnσ = 0.40, a total of 144 groups of input. At the same time, in order to verify the improvement of the inversion stability of the method proposed in this invention, this embodiment selects the microphysical characteristics simulation input as m r =1.45,m i =0.005, r med =125nm, lnσ=0.40 and the inversion was repeated 100 times.

[0083] Step 2: Based on the k-nearest neighbor theory, the gap between the normalized input and each group of normalized optical properties in the normalized lookup table is calculated and sorted from small to large, and the elements of the groups with the smallest gap are retained to obtain a narrowed solution space;

[0084] Step 2 calculates the difference between the normalized input and each set of normalized optical properties in the normalized lookup table as follows:

[0085]

[0086] Among them, Dist M (n) represents the difference between each set of normalized optical properties in the normalized lookup table and the normalized input, represents the normalized optical properties in the normalized lookup table, n represents the nth group element, G input represents the normalized input, S represents With G input The covariance matrix of

[0087] Step 2 sorts the elements in ascending order and retains the elements with the smallest gap, as follows:

[0088] N kNN =ω1*N LUT

[0089] where N kNNrepresents the number of element groups retained, and ω1 represents the ratio of the number of groups retained when generating the reduced solution space; in this embodiment, ω1=0.01.

[0090] Step 3: Calculate the gap between the normalized input and each normalized optical property in the reduced solution space, sort the normalized optical properties from large to small based on the gap between the normalized input and each normalized optical property in the reduced solution space, and generate decision trees with different pruning orders based on the weighted sorting. Prune each decision tree according to the included pruning order to generate possible solutions. Average the possible solutions of all decision trees to obtain the first set of microphysical property solutions.

[0091] The calculation of the normalized input described in step 3 and the gap between each normalized optical property in the solution space are as follows:

[0092]

[0093] Among them, Dist opt (i) represents the gap between the normalized input and each normalized optical property in the reduced solution space, represents the normalized optical properties in the reduced solution space, i represents different normalized optical properties;

[0094] In step 3, the normalized optical properties are sorted from large to small based on the gap between the normalized input and each normalized optical property in the solution space, and a decision tree with different pruning orders is generated based on the weighted sorting, as follows:

[0095]

[0096] Among them, P(i) represents the probability of the i-th normalized optical property being extracted after sorting; the process of generating decision trees with different pruning orders is: the total number of generated trees is N RF For each decision tree, all normalized optical properties are extracted without replacement according to P(i) to generate a sequence of normalized optical properties as the pruning order;

[0097] Each decision tree described in step 3 is pruned according to the included pruning order to generate possible solutions, as follows:

[0098] N t (i+1)=N t (i)×(1-ω2)

[0099] 0≤i≤N norm -1,N t (0) = N kNN

[0100] where N t(i) represents the number of element groups retained after the i-th pruning, ω2 represents the pruning coefficient, N t (0) represents the initial number of element groups; in this embodiment, N RF =500,ω2=0.35.

[0101] Step 4: Calculate the corresponding optical properties based on the first set of microphysical property solutions and normalize them to obtain the first set of optical property solutions. Based on the k-nearest neighbor theory, calculate the gap between the union of the first set of microphysical property solutions and the first set of optical property solutions and each element in the normalized lookup table, and retain the elements in the groups with the smallest gap to obtain the constraint window.

[0102] As described in step 4, based on the k-nearest neighbor theory, the gap between the union of the first set of microphysical property solutions and the first set of optical property solutions and each element of the normalized lookup table is calculated as follows:

[0103]

[0104] Among them, Dist M,ret (n) represents the difference between the first set of microphysical property solutions and the first set of optical property solutions and each element of the normalized lookup table, G LUT (n) represents the normalized lookup table, n represents the nth group element of the normalized lookup table, G ret S represents the union of the first set of microphysical property solutions and the first set of optical property solutions G LUT (n) and G ret The covariance matrix of

[0105] Step 4 sorts the elements in ascending order and retains the elements with the smallest gap, as follows:

[0106] N C =ω3*N LUT

[0107] where N C represents the number of element groups retained, and ω3 represents the ratio of the number of groups retained by the generated constraint window; in this embodiment, ω3 = 0.01

[0108] Step 5: Based on the k-nearest neighbor theory, follow the method in step 2 and use the constraint window instead of the normalized lookup table. Reduce the constraint window and perform interpolation to obtain the refined solution space.

[0109] The interpolation operation is performed on the reduced constraint window as described in step 5, as follows:

[0110] First, three of the real part of the complex refractive index, the imaginary part of the complex refractive index, the mode radius, and the logarithmic geometric variance are fixed, and all values ​​of the remaining microphysical properties and their corresponding optical properties are extracted in the reduced constraint window, a total of Nnorm The same processing is performed on the remaining three of the real part of the complex refractive index, the imaginary part of the complex refractive index, the mode radius and the logarithmic geometric variance to obtain the refined solution space; define P inter is the interpolation ratio, each P inter Interpolation is performed between data points; in this embodiment, P inter =2.

[0111] Step 6: Following the method of step 3, the refined solution space is used instead of the reduced solution space to obtain the final solution of the microphysical properties and the final solution of the optical properties, and the remaining microphysical properties are calculated;

[0112] The remaining microphysical properties are calculated as described in step 6, as follows:

[0113]

[0114]

[0115]

[0116]

[0117] in, represents the volume concentration obtained by inversion, represents the backscattering coefficient at the input k-th backscattering wavelength, represents the extinction coefficient at the input nth extinction wavelength, represents the backscattering coefficient at the kth backscattering wavelength obtained by inversion, Indicates the extinction coefficient at the nth extinction wavelength obtained by inversion, which can be calculated according to the method in step 1, N λ is the sum of the backscattering wavelength and the extinction wavelength;

[0118] v ret (r) represents the volume particle size spectrum distribution obtained by inversion, lnσ ret represents the logarithmic geometric variance obtained by inversion, represents the pattern radius obtained by inversion, represents the effective particle size obtained by inversion, Indicates the inverted wavelength λ x The single scattering albedo at , represents the real part of the complex refractive index obtained by inversion, represents the imaginary part of the complex refractive index obtained by inversion, and Represents wavelength λ x The scattering coefficient and extinction coefficient at can be calculated by Mie theory; in this embodiment, λx =532nm.

[0119] like Figure 2 As shown in Figure 2, (a)-(d) show the absolute errors of the real and imaginary parts of the complex refractive index, and the relative errors of the effective particle size and volume concentration, respectively, for the inversion of the test data set using the 3β+2α, 3β+1α, and 2β+1α configurations. Note that the inversion errors for all three configurations are small, indicating reliable inversion accuracy. Furthermore, there is no significant difference in the performance between the three configurations, indicating that this method can be effectively applied to simpler lidar systems, which is of great significance for reducing the cost and expanding the application scope of multi-wavelength lidar.

[0120] like Figure 3 Figures (a)-(d) show boxplots of 100 repeated inversions using selected simulated microphysical properties in a 3β+2α configuration. The "basic method" in the figure represents a method that includes only steps S1, S2, and S3 and does not apply a weighted decision tree generation scheme, which is compared with our method. Note that our method significantly improves the inversion stability of microphysical properties and exhibits high reliability.

[0121] An embodiment of the system of the present invention is a multi-wavelength lidar aerosol microphysical property inversion system, comprising:

[0122] The normalized lookup table and normalized input construction modules are used to calculate the backscattering coefficient at each backscattering wavelength and the extinction coefficient at each extinction wavelength based on the Mie theory using the aerosol lognormal particle size distribution model. The normalized lookup table is then constructed using the normalization method to obtain the normalized input.

[0123] A solution space reduction calculation module is used to calculate the gap between the normalized input and each group of normalized optical properties in the normalized lookup table based on the k-nearest neighbor theory, sort them from small to large, and retain the group elements with the smallest gap to obtain a reduced solution space;

[0124] A first set of microphysical property solution solving modules is configured to calculate the difference between the normalized input and each normalized optical property in the reduced solution space, sort the normalized optical properties from largest to smallest based on the difference between the normalized input and each normalized optical property in the reduced solution space, generate decision trees with different pruning orders based on the weighted sorting, prune each decision tree according to the included pruning order, generate possible solutions, and average the possible solutions of all decision trees to obtain the first set of microphysical property solutions;

[0125] A constraint window construction module is used to calculate the corresponding optical properties based on the first set of microphysical property solutions and normalize them to obtain the first set of optical property solutions. Based on the k-nearest neighbor theory, the module calculates the gap between the union of the first set of microphysical property solutions and the first set of optical property solutions and each set of elements in the normalized lookup table, and retains the elements of the groups with the smallest gap to obtain the constraint window.

[0126] A refined solution space construction module is used to obtain a refined solution space based on the k-nearest neighbor theory, through the first set of microphysical characteristic solution solving modules, using a constraint window instead of a normalized lookup table, reducing the constraint window, and performing interpolation operations;

[0127] The final solution solving module, combined with the first set of microphysical characteristic solution solving modules, uses the refined solution space instead of the reduced solution space to obtain the final solution of the microphysical characteristics and the final solution of the optical characteristics, and calculates the remaining microphysical characteristics.

[0128] The normalized lookup table, normalized input construction module, solution space reduction calculation module, first group of microphysical characteristic solution solving module, constraint window construction module, refined solution space construction module, and final solution solving module are all deployed on the server.

[0129] The foregoing is merely a list of specific embodiments of the present application, intended to enable those skilled in the art to understand or implement the present application. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present application. Therefore, the present application is not limited to the embodiments shown herein, but is intended to conform to the broadest scope consistent with the principles and novel features claimed herein.

Claims

1. A multi-wavelength lidar aerosol microphysical properties inversion method, characterized by: The following steps are involved: Step 1: Based on the Mie theory, the aerosol lognormal particle size distribution model is used to calculate the backscattering coefficient at each backscattering wavelength and the extinction coefficient at each extinction wavelength. A lookup table is then established. The optical characteristics of the lookup table and the input of the multi-wavelength lidar are normalized using a normalization method to obtain a normalized lookup table and normalized input. Step 2: Based on the k-nearest neighbor theory, the gap between the normalized input and each group of normalized optical properties in the normalized lookup table is calculated and sorted from small to large, and the elements of the groups with the smallest gap are retained to obtain a narrowed solution space; Step 3: Calculate the gap between the normalized input and each normalized optical property in the reduced solution space, sort the normalized optical properties from large to small based on the gap between the normalized input and each normalized optical property in the reduced solution space, and generate decision trees with different pruning orders based on the weighted sorting. Prune each decision tree according to the included pruning order to generate possible solutions. Average the possible solutions of all decision trees to obtain the first set of microphysical property solutions. Step 4: Calculate the corresponding optical properties based on the first set of microphysical property solutions and normalize them to obtain the first set of optical property solutions. Based on the k-nearest neighbor theory, calculate the gap between the union of the first set of microphysical property solutions and the first set of optical property solutions and each element in the normalized lookup table, and retain the elements in the groups with the smallest gap to obtain the constraint window. Step 5: Based on the k-nearest neighbor theory, follow the method in step 2 and use the constraint window instead of the normalized lookup table. Reduce the constraint window and perform interpolation to obtain the refined solution space. Step 6: Following the method of step 3, the refined solution space is used instead of the reduced solution space to obtain the final solution of the microphysical properties and the final solution of the optical properties, and calculate the remaining microphysical properties.

2. The multi-wavelength lidar aerosol microphysical property inversion method according to claim 1, characterized in that: According to the Mie theory described in step 1, the aerosol log-normal particle size distribution model is used to calculate the backscattering coefficient at each backscattering wavelength and the extinction coefficient at each extinction wavelength and establish a lookup table, as follows: Input the real part of the complex refractive index, the imaginary part of the complex refractive index, the mode radius and the logarithmic geometric variance in the aerosol microphysical properties, and calculate in sequence the backscattering coefficient at each backscattering wavelength according to the Mie theory through the aerosol lognormal particle size distribution model according to each backscattering wavelength. Combined with each extinction wavelength, calculate in sequence the extinction coefficient at each extinction wavelength according to the Mie theory through the aerosol lognormal particle size distribution model; according to the number of groups N LUT The real part of the complex refractive index, the imaginary part of the complex refractive index, the mode radius and the logarithmic geometric variance are used to calculate the corresponding backscattering coefficient and extinction coefficient, and a lookup table is constructed together; the lookup table contains N LUT Group elements, each group includes the real part of the complex refractive index, the imaginary part of the complex refractive index, the mode radius and the logarithmic geometric variance as well as the corresponding backscattering coefficient and extinction coefficient.

3. The multi-wavelength lidar aerosol microphysical property inversion method according to claim 2, characterized in that: The combined normalization method is used to normalize the lookup table optical characteristics and the input of the multi-wavelength lidar, as follows: N norm =N+K+N×K Among them, β k represents the backscattering coefficient at the kth backscattering wavelength, B k represents the normalized backscattering coefficient at the kth backscattering wavelength, B norm represents the second-order norm of the backscattering coefficient at all backscattering wavelengths, K represents the number of backscattering wavelengths; α n represents the extinction coefficient at the nth extinction wavelength, A n represents the normalized extinction coefficient at the nth extinction wavelength, A norm represents the second-order norm of the optical coefficient at all extinction wavelengths, and N represents the number of extinction wavelengths; Represents the extinction backscattering ratio between the kth backscattering wavelength and the nth extinction wavelength; the normalized optical properties include the normalized backscattering coefficient, the normalized extinction coefficient and the extinction backscattering ratio, N norm Represents the normalized optical property quantity.

4. The multi-wavelength lidar aerosol microphysical property inversion method according to claim 3, characterized in that: Step 2 calculates the difference between the normalized input and each set of normalized optical properties in the normalized lookup table as follows: Among them, Dist M (n) represents the difference between each set of normalized optical properties in the normalized lookup table and the normalized input, represents the normalized optical properties in the normalized lookup table, n represents the nth group element, G input represents the normalized input, S represents With G input The covariance matrix of Step 2 sorts the elements in ascending order and retains the elements with the smallest gap, as follows: N kNN =ω1*N LUT where N kNN Represents the number of element groups retained, and ω1 represents the ratio of the number of groups retained when generating the reduced solution space.

5. The multi-wavelength lidar aerosol microphysical property inversion method according to claim 4, characterized in that: The calculation of the normalized input described in step 3 and the gap between each normalized optical property in the solution space are as follows: Among them, Dist opt (i) represents the gap between the normalized input and each normalized optical property in the reduced solution space, represents the normalized optical properties in the reduced solution space, i represents different normalized optical properties; In step 3, the normalized optical properties are sorted from large to small based on the gap between the normalized input and each normalized optical property in the solution space, and a decision tree with different pruning orders is generated based on the weighted sorting, as follows: Among them, P(i) represents the probability of extracting the i-th normalized optical property after sorting; the process of generating decision trees with different pruning orders is: the total number of generated trees is N RF For each decision tree, all normalized optical properties are extracted without replacement according to P(i) to generate a sequence of normalized optical properties as the pruning order; Each decision tree described in step 3 is pruned according to the included pruning order to generate possible solutions, as follows: N t (i+1)=N t (i)×(1-ω2) 0≤i≤N norm -1,N t (0)=N kNN where N t (i) represents the number of element groups retained after the i-th pruning, ω2 represents the pruning coefficient, N t (0) represents the initial number of element groups.

6. The multi-wavelength lidar aerosol microphysical property inversion method according to claim 5, characterized in that: As described in step 4, based on the k-nearest neighbor theory, the gap between the union of the first set of microphysical property solutions and the first set of optical property solutions and each element of the normalized lookup table is calculated as follows: Among them, Dist M,ret (n) represents the difference between the first set of microphysical property solutions and the first set of optical property solutions and each element of the normalized lookup table, G LUT (n) represents the normalized lookup table, n represents the nth group element of the normalized lookup table, G ret S represents the union of the first set of microphysical property solutions and the first set of optical property solutions G LUT (n) and G ret The covariance matrix of Step 4 sorts the elements in ascending order and retains the elements with the smallest gap, as follows: N C =ω3*N LUT where N C Represents the number of element groups retained, and ω3 represents the ratio of the number of groups retained by the generated constraint window.

7. The multi-wavelength lidar aerosol microphysical property inversion method according to claim 6, characterized in that: Step 5 performs interpolation processing on the reduced constraint window, as follows: Fixing three of the real part of the complex refractive index, the imaginary part of the complex refractive index, the mode radius and the logarithmic geometric variance, extract all the values ​​of the remaining microphysical properties and their corresponding optical properties in the reduced constraint window, a total of N norm The same processing is performed on the remaining three of the real part of the complex refractive index, the imaginary part of the complex refractive index, the mode radius and the logarithmic geometric variance to obtain the refined solution space; define P inter is the interpolation ratio, each P inter Interpolate between data points.

8. The multi-wavelength lidar aerosol microphysical property inversion method according to claim 7, characterized in that: The remaining microphysical properties are calculated as described in step 6, as follows: in, represents the volume concentration obtained by inversion, represents the backscattering coefficient at the input k-th backscattering wavelength, represents the extinction coefficient at the input nth extinction wavelength, represents the backscattering coefficient at the kth backscattering wavelength obtained by inversion, represents the extinction coefficient at the nth extinction wavelength obtained by inversion, which is calculated according to the method in step 1. is the sum of the backscattering wavelength and the extinction wavelength; v ret (r) represents the volume particle size spectrum distribution obtained by inversion, lnσ ret represents the logarithmic geometric variance obtained by inversion, represents the pattern radius obtained by inversion, represents the effective particle size obtained by inversion, Indicates the inverted wavelength λ x The single scattering albedo at , represents the real part of the complex refractive index obtained by inversion, represents the imaginary part of the complex refractive index obtained by inversion, and Represents wavelength λ x The scattering coefficient and extinction coefficient at can be calculated by Mie theory.

9. A multi-wavelength lidar aerosol microphysical properties inversion system, characterized by: include: The normalized lookup table and normalized input construction modules are used to calculate the backscattering coefficient at each backscattering wavelength and the extinction coefficient at each extinction wavelength based on the Mie theory using the aerosol lognormal particle size distribution model, and then construct a lookup table. The normalization method is combined to normalize the optical characteristics of the lookup table and the input of the multi-wavelength lidar to obtain the normalized lookup table and normalized input. A solution space reduction calculation module is used to calculate the gap between the normalized input and each group of normalized optical properties in the normalized lookup table based on the k-nearest neighbor theory, sort them from small to large, and retain the group elements with the smallest gap to obtain a reduced solution space; A first set of microphysical property solution solving modules is configured to calculate the difference between the normalized input and each normalized optical property in the reduced solution space, sort the normalized optical properties from largest to smallest based on the difference between the normalized input and each normalized optical property in the reduced solution space, generate decision trees with different pruning orders based on the weighted sorting, prune each decision tree according to the included pruning order, generate possible solutions, and average the possible solutions of all decision trees to obtain the first set of microphysical property solutions; A constraint window construction module is used to calculate the corresponding optical properties based on the first set of microphysical property solutions and normalize them to obtain the first set of optical property solutions. Based on the k-nearest neighbor theory, the module calculates the gap between the union of the first set of microphysical property solutions and the first set of optical property solutions and each set of elements in the normalized lookup table, and retains the elements of the groups with the smallest gap to obtain the constraint window. A refined solution space construction module is used to obtain a refined solution space based on the k-nearest neighbor theory, through the first set of microphysical characteristic solution solving modules, using a constraint window instead of a normalized lookup table, reducing the constraint window, and performing interpolation operations; The final solution solving module, combined with the first set of microphysical characteristic solution solving modules, uses the refined solution space instead of the reduced solution space to obtain the final solution of the microphysical characteristics and the final solution of the optical characteristics, and calculates the remaining microphysical characteristics.

Images